Descriptive Statistics
Table 1 presents the descriptive statistics and correlations of the measures. The table shows that, on average, analysts’ forecasts of FFO tend to be higher than actual FFO announced by REITs. Analysts’ forecasts of net income (NI) also tend to be higher than the actual. Correlation between actual FFO and analyst estimates of FFO is 0.90 indicating that higher estimates are followed by higher actual FFO. The correlation between actual NI and analyst estimates is only 0.565 indicating that higher estimates of NI are not necessarily followed by higher actual NI. The mean number of analysts forecasting FFO and NI is 8 and 3 respectively. The fact that more analysts provide estimates of FFO than NI suggests that investors have a greater demand for FFO.
FFO Announcements and Abnormal Trading Volume
Tables 2 and 3 present the abnormal trading volume generated on the announcement of FFO by REITs based on two alternative definitions. As the results show, for all windows used, there is significant abnormal trading volume around the announcement dates. This suggests that the market engages in significantly more trading when REITs announce FFO irrespective of whether it misses, meets or beats the consensus forecast by analysts. The table also shows that the level of trading prior to the announcement of FFO is negative and significant in the market-based specifications suggesting that the market tends to reduce trading activity in anticipation of the FFO announcement. The pattern of abnormal trading volume prior to and around the date of the
- 32 -
announcement of FFO is consistent with the trading pattern around the announcement of earnings information for non-REITs. (See for example Chae, 2005)
To examine whether the level of abnormal trading volume is associated with the level of the surprise contained in the FFO announcement, I estimate three regressions. The first is based on the traditional specifications frequently used in the existing literature that uses the difference between actual FFO and expected FFO scaled by price as an independent variable to explain abnormal trading volume. The second specification modifies the first by allowing for different slopes for positive and negative surprises and use piecewise regression. Lastly, I use an unconstrained polynomial regression that allows for curvilinearity in the relation between abnormal trading volume and FFO surprises.
Table 4 presents the results based on the traditional specification. The results show that for all windows, there is no significant relationship between abnormal trading volume and the level of surprise in the announcement of FFO. This is indicated by R2s that are not statistically different from zero. Since the traditional specification is a constrained version of the specification that uses both expected and actual FFO as independent variables, I present an unconstrained linear regression in Table 4. This allows me to formally test whether the constraints imposed by the traditional specification may be driving the results. The result of the F-test shows no significant difference in the variance explained by the traditional specification and the unconstrained version.
- 33 -
The results from the piecewise regression are presented in Table 5. The results show no significant relationship as indicated by the R2s that are not statistically different from zero. In addition, tests of the difference in model R2s show no significant difference between the piecewise regression and the traditional specifications. This suggests that the lack of association between abnormal trading volume and the size of the surprise holds when I allow for different slopes for negative and positive surprises. This is further confirmed by the insignificant coefficient on the coded variable.
I present the results based on the unconstrained polynomial regression in Table 6. The results show insignificant R2s for all windows. An F-test for difference in R2s for the unconstrained linear and quadratic specifications shows no significant difference. These results suggest that the lack of significant relationship between abnormal trading volume and the size of surprise is robust to different specifications.
These results are inconsistent with the findings documented for non-REITs that also use trading volume as a measure of investor response (see Bamber, 1987, Bamber and Cheon, 1995 and Cready and Hurtt, 2002).Moreover, results from a number of studies examining the information content of FFO using abnormal returns as a gauge of investor reaction have documented significant relationship between FFO surprises and abnormal returns (see Gyamfi-Yeboah, Ziobrowski and Lambert, 2010 and Baik, Billings and Morton 2008). This suggests that there may be peculiar characteristics of REITs that limits investors’ ability to fully trade on the announcement of FFO surprises. Barclay, Kandel and Marx (1998) show that higher transaction
- 34 -
costs significantly reduce trading volume but have no effect on share prices. This suggests that in the presence of higher transaction costs, market response through share price changes could be more pronounced than changes in trading volume. It is well documented (Subrahmanyam, 2007; Bertin, Kofman, Michayluk and Prather, 2005 and Ghosh, Miles and Sirmans, 1996) that REITs have, on average, a relatively high bid-ask spread (a proxy for transaction cost) and are relatively less liquid compared to non-REITs. It is plausible to argue that the higher bid-ask spreads for REITs, which reflects a lack on consensus on prices may limit the amount of trading that occurs in response to the announcement of FFO surprises.
Information Content of Net Income vs. FFO (Test of Hypothesis 3)
To address the question of whether FFO conveys more useful information about REITs performance when compared to traditional GAAP measures, notably net income, I re-estimate all equations using net income in place of FFO. Since the evidence presented in both Downs and Guner (2006) and Baik, Billings and Morton (2008) suggest that FFO may be more useful to investors, I expect the variance explained by the regressions containing FFO to be significantly larger than those containing net income.
I present the results of the FFO versus Net Income regressions in Table 7. Since the preceding results indicate no significant relation between abnormal trading volume and FFO surprises, I use abnormal returns as the dependent variable. To ensure that the results are robust, I use two specifications: the constrained linear regression (difference score) and the unconstrained
- 35 -
polynomial regression. As the results show, R2 from the regression using net income is substantially lower than the R2 from the regression using FFO. A test for the difference in R2 using Clarke’s distribution free test shows that FFO explains significantly more variance in abnormal returns than net income supporting our conjecture that FFO provides more useful information to investors than net income.
Informed Traders and Abnormal Trading Volume (Test of Hypothesis 4)
I test Kim and Verrecchia’s (1991) hypothesis that firms with more informed traders will experience less abnormal trading on the announcement of earnings (or FFO). The results of the regression testing this hypothesis are presented in Table 8. The main variable of interest is the proportion of outstanding shares held by institutions (IO) at the end of each quarter prior to the FFO announcement. I include control variables that have been documented in prior studies to be significantly related to abnormal trading volume. I find, as expected, that absolute abnormal returns are significantly positively related to abnormal trading volume. I also find price to be significantly related to the abnormal trading while firm size does not appear to have any significant impact. As the results in the table show, even though the proxy for the level of informed traders enters the regression with the expected sign, the coefficient is not statistically different from zero. Consistent with the approach adopted in prior studies, I include a quadratic term for IO but obtain the same results. This suggests that, for REITs, the level of institutional ownership does not appear to have a significant impact on the abnormal trading volume observed on the announcement of FFO. This result is in contrast to the findings documented in Kim, Krinsky and Lee (1997) and Utama and Cready (1997) for non-REITs. I speculate that the
- 36 -
relatively high transaction costs for REITs may limit the trading by uninformed traders who may want to trade on the announcement of FFO.
To formally test whether the results documented above are driven by the relatively higher transaction costs for REITs, I interact log price (a proxy for transaction cost) with IO and include this term in a regression predicting abnormal trading volume. The results are presented in Table 9. A significant and positive coefficient on the interaction term would indicate that the impact of the level of institutional ownership on abnormal trading volume depends on the level of transaction costs. Specifically, such a result would suggest that REITs with lower transaction costs and higher levels of institutional ownership are more likely to have lower volume of trading on the announcement of FFO. As the results show, the coefficient on the interaction term is positive and significant supporting the conjecture that the lack of significance on the institutional ownership variable may partly be explained by the relatively high transaction costs for REITs. Figure 2 demonstrates the relation between abnormal trading volume and the levels of institutional ownership at three different values of price: the mean, one standard deviation above the mean and below the mean. The y-axis shows cumulative abnormal trading volume while the x-axis shows the centered values for levels of institutional ownership. As the figure shows, the slope for the level of institutional ownership varies for different values of price.
Dennis and Strickland (2002) document different trading behavior for different types of institutional investors. The authors decompose institutional ownership into four categories and find some classes of institutional investors, especially mutual funds and investment advisors, to
- 37 -
be more active traders than other classes such as banks and insurance companies. They argue that mutual funds and investment advisors tend to “herd together and trade with the momentum”. It is important to point out that after 1998, the break down of institutional ownership into the four categories is unreliable due to coding errors. As a result, the purpose of this analysis is to assess whether the use of an aggregate measure of institutional ownership is appropriate and not so much on the trading behavior of each category of institutional investors. Figure 3 shows the proportion of institutional ownership in REITs by institution type over the sample period. The average institutional ownership over the sample is about 73%. The pattern of institutional ownership is stable over the sample period with no discernable shifts in ownership patterns.
Table 10 contains the results of the analysis using two broad classifications of institutional investors: namely (1) mutual funds and investment advisor (IOM) and (2) all other institutions (IOA). The classification is based on the coding used in the Thompson Reuters 13(f) data. The decision to form the two groups was influenced by the findings in Dennis and Strickland (2002), which suggest that mutual funds and investment advisors may trade frequently and have a shorter holding period than pension funds, banks and insurance companies. I expect the coefficient on the IOM variable to be positive and significant while that on IOA is expected to be negative and significant. As the results show, there is a significant and positive relationship between the level of holdings by mutual funds and investment advisors and the level of abnormal trading volume around FFO announcement. I also find a significant negative relationship at the 10% level between the level of ownership by other institutional investors (pension funds & endowments, banks and insurance) and the level of abnormal trading volume around the announcement of
- 38 -
FFO. Taken together, these and the previous results using aggregated institutional ownership variable suggest that institutional investors are not homogeneous and may therefore exhibit different trading behaviors. The results also indicate that the use of an aggregate measure of institutional ownership as proxy for informed traders to test hypotheses such as Kim and Verracchia’s has the potential of masking the predicted relationships. These results bring into question the appropriateness of using aggregate institutional ownership as a proxy for informed traders since as Downs and Guner (1999) note investors may trade for reasons other than informational advantage.
Robustness checks
To assess the robustness of the results, I carry out analyses based on FFO announcements between 1997 and 1999. Tables 11 and 12 contain the cumulative abnormal trading volume around the announcement of FFO. In contrast to the results for 200 -2006 period, investors appear not to engage in significantly more trading on the announcement of FFO except for positive surprises when abnormal trading volume is measured based on firm specific data. The market-based measure is, however, generally consistent with the 2004 -2006 results. To examine whether the level of abnormal trading volume is related to the surprise contained in the FFO announcement, I use three different specifications. The results are presented in Tables 13, 14, 15 for the difference score, piecewise regression and polynomial regression specifications respectively. In all three specifications, there is no significant relationship between the level of abnormal trading volume around the announcement of FFO and the surprise contained in the announcement. Additional analyses (for both 2004 -2006 and 1997 – 1999 sample periods) that
- 39 -
include an interaction variable created between the time dummies and FFO surprises show no significant time variation in the relation between cumulative abnormal trading volume and FFO surprises.
I also examine the impact informed traders have on the levels of abnormal trading volume for the 1997 -1999 sample period. Figure 4 shows the proportion of REITs shares held by the different categories of institutional investors. On average, institutions held about 52% of REIT outstanding shares over this period. Note that the pattern of institutional ownership by category experienced a significant shift during the first quarter of 1999. Prior to this time, mutual funds and investment advisors held a significant majority of REIT shares. Even though the total number of shares held by institutions in the aggregate did not significantly change after the first quarter of 1999, mutual funds and investment advisors significantly reduced their stakes in REITs while pension funds and endowment appear to have taken up the shares previously held by mutual funds and investment advisors. The apparent shift in ownership among institutions may be the result of coding errors that occurred around this time. As a result of this shift in ownership among institutions, I create an indicator variable to control for any impact the shift in ownership pattern may have on the results.
The first set of results, presented in Table 16, examines how institutions in the aggregate impact the level of abnormal trading volume. There is no significant relationship between the proportion of shares held by institutions and abnormal of trading volume on the announcement of FFO confirming the earlier results. The results, using two broad classifications of institutional investors: namely mutual funds and investment advisors and all other institutions are presented
- 40 -
in Table 17. The results indicate no significant relationship between any of the institutional categories and abnormal trading volume. To control for the shift in ownership that occurred in the first quarter of 1999, I create an indicator variable, which is coded 1 for ownership post 1999 first quarter and 0 otherwise. I then create an interaction variable between the two categories of ownership and the indicator variable.
The results, which are presented in Table 18, show that even though there is a significant positive relationship between holdings by other institutions (pension funds, banks and insurance), the relationship is significantly lower after 1999. The coefficient on ownership by mutual funds and investment advisors is not significant. These results, even though in contrast to those for the 2004 to 2006 sample period4, suggest that institutional investors are heterogeneous in their trading behaviors. It may therefore be inappropriate to use an aggregate measure of institutional ownership in examining the trading behavior of institutions.
Lastly, to examine whether the coefficient on institutional ownership is time varying I interact the time dummies with the institutional ownership variable. The results of the regression analysis show no significant relation between the interaction variable and cumulative abnormal trading volume. This suggests that the lack of association between the levels of institutional ownership (in the aggregate) and abnormal trading volume around the announcement of FFO is not time varying.
- 41 -
Dispersions in Analysts’ Forecast and Abnormal Trading Volume
In addition to using the proportion of institutional ownership as a proxy for the number of informed traders to test Kim and Verracchia’s hypothesis, a number of researchers have examined how other measures of predisclosure uncertainty affect the level of abnormal trading volume around the announcement of earnings (see Bamber, Barron and Stober, 1997, Atiase and Bamber, 1994). A measure that has frequently been used in the literature is the dispersion in analysts’ forecast prior to the announcement. But as already noted in the literature review section, this measure may reflect divergent expectations among analysts and not necessarily differences in the quality of predisclosure information as the theoretical proposition predicts. Bamber, Barron and Stober, (1997) drawing support from Barron, Kim, Lim and Stevens (1997), argue that Kim and Verracchia’s model supports the conjecture that trading volume will be positively related to dispersion in analysts’ forecasts.
Notwithstanding this limitation, the dispersion in analysts’ forecasts has been found to be positively related to the level of abnormal trading volume (Bamber, Barron and Stober, 1997, Atiase and Bamber, 1994). This suggests that higher levels of disagreement among analysts should result in greater levels of trading on the announcement of earnings (FFO). As a further test of robustness, I regress the abnormal trading volume on the dispersion in analysts’ forecast and report the results in Table 19. There is a significant and positive relationship between the level of dispersion in analysts’ forecasts and the level of abnormal trading volume around the announcement of FFO. The coefficients on the control variables are generally as expected and documented in our previous results using the proportion of institutional ownership.
- 42 -
CHAPTER 5